Appointed O cials and Consolidation of New Democracies: Evidence from Indonesia

Appointed O cials and Consolidation of New Democracies: Evidence from Indonesia Monica Martinez-Bravo MIT January 15th, 2010 JOB MARKET PAPER Abstract The workings of new democracies are heavily in uenced by the legacies of the previous autocratic regimes. This paper examines the e ect of one of these legacies on the process of democratic consolidation: the body of appointed o cials at the lowest level of the administration. My theoretical analysis highlights that appointed o cials have a vested interest in the persistence of the autocratic status quo in order to protect their jobs. At the onset of the rst democratic election, they use local patronage networks to promote the electoral chances of the dictator s party. However, if there is imperfect information about the political leanings of local o cials and if the reformist party is expected to win by a large majority, this e ect can be reversed: opportunistic local o cials will, in that case, pretend to be strong supporters of the reformist party, in an attempt to keep their jobs. I test the empirical predictions of the model with a unique data set containing information on the electoral results for the rst and second post-suharto elections for over 30,000 villages in Indonesia. Within districts, Suharto s party was 5 percentage points more likely to win in villages with an appointed village head than in those with an elected village head. The results are robust to the inclusion of a wide set of controls and similar across econometric methods (OLS, propensity score matching). Consistent with the model, this e ect is only reversed for districts in which the reformist party won by a large margin. Overall, this paper provides substantial evidence that, unless reformist parties are expected to be the clear winners of the rst democratic elections, appointed o cials will promote the electoral chances of the dictator s party, which could become an impediment for the process of democratic consolidation. I would like to thank my advisors, Daron Acemoglu, Abhijit Banerjee, and Ben Olken, for their outstanding advice and encouragement throughout this project. I would also like to thank Jim Snyder, Gerard Padró-i-Miquel, Esther Du o, Eric Weese, Dave Donaldson, Allan Drazen, Nancy Qian, Mar Reguant, Martí Mestieri, Xavier Giné, Julia Tobias, Kevin Evans, Scott Guggenheim, Hans Antlöv, Cynthia Kinnan, Horacio Larreguy, Patrick Warren, Pablo Querubin, and Leopoldo Fergusson, for valuable comments and discussion. I am grateful to Fundación Ramón Areces for their nancial support. 1

1 Introduction The rst years of a democratic regime are when democracy is most vulnerable. Many scholars have recognized that several practices or institutions developed during the previous autocratic regime leave a legacy that will condition the workings of the new democracy, both in terms of their economic and political outcomes. 1 A crucial juncture when these legacies might play a critical role is at the time of the rst democratic election. If the presence of these legacies prevents reformist parties from taking o ce, democratic deepening reforms might not be implemented, what could lead to a captured or unconsolidated democracy. This paper examines, theoretically and empirically, how one of these legacies can a ect the outcome of the rst democratic election: the body of local o cials. There is extensive evidence that documents the importance of local leaders for many nondemocratic regimes. 2 By means of local patronage networks and other intimidation mechanisms, they are able to obtain support for the regime, or simply compliance, from the population. At the onset of the rst democratic election, most of them will still be in their positions and still have the means and the ability to in uence voters. However, we lack a good understanding of what incentives local o cials face to continue to use these local patronage networks in the rst democratic election. First, the loyalty ties that local o cials had with the previous autocratic regime can suddenly change, given the new political scenario. Second, new democracies are characterized by a great amount of uncertainty regarding real political leanings. The repressive nature of nondemocratic regimes prevents the disclosure of political views di erent from those of the dictator s ideology. Consequently, the high degree of asymmetric information with which new democracies are endowed enables local o cials to behave opportunistically during the early stages of the democratic period. This unique political environment raises a number of questions: What incentives does local o cials face to continue to in uence voters in the rst democratic election? Will they use the patronage network to obtain support for the previous dictator s party? Will they ever give their support to reformist parties? This paper examines these questions and also explores how the answers depend on the method of selection of local o cials, in particular, on whether local o cials are appointed by some upper level government, or elected in local elections. In order to provide answers for these questions, I develop a model in which two parties (the dictator s party and a reformist party) contest the rst democratic election for some upper-level o ce. At the lower level, there are local o cials that control the patronage networks and decide how much e ort to exert during the electoral campaign to in uence voters in their region. Since the 1 See, for instance, Acemoglu, Ticchi and Vindigni (2008, 2009), for how the legacy of an ine cient bureaucratic structure or a large military might a ect politics in transitional democracies. 2 Some examples are Baum and Shvchenko (1999) discussing the case of China, Magaloni (2006) on Mexico, Pepinsky (2007) on Malaysia, and Blaydes (2008) on Egypt. 2

regime has just transitioned from a nondemocratic period, there is imperfect information about their political leanings. Hence, e ort has a twofold motivation in this model: it can signal a particular political leaning to upper levels of government and it can also in uence the outcome of the election. The model highlights that local o cials who are appointed by upper levels of government have a much stronger incentive to in uence voters in their region. They do so because they will be able to keep their jobs only if the party they support wins the election and the winner of the election is con dent enough that the local o cial is truly one of her supporters. In contrast, elected local o cials lack this incentive, because the continuity in their positions does not depend on the outcome of upper-level elections: they were elected into o ce through local elections and they will remain in o ce until local elections are held again. The decision problem of appointed local o cials constitutes a signaling game that has two broad types of equilibria: pooling and separating. 3 The model predicts that, if the election is expected to be lopsided, a pooling equilibrium emerges in which all appointed o cials exert the same level of e ort, regardless of their real political leanings. This is indeed a very intuitive result: when, ex-ante, one of the candidates is very likely to win, all the appointed o cials exert e ort to support that candidate and pretend to be her strongest supporters. If the election is expected to be contested, a separating equilibrium emerges: in the absence of a clear winner, each appointed o cial nds it optimal to support his most preferred candidate. In separating equilibria, since appointed o cials are exerting e ort in opposite directions, the net e ect depends on the proportion of them that are supporters of each party. However, the likely higher proportion of dictator s supporters would tilt the balance towards the dictator s party. Therefore, we would expect that in most scenarios, appointed o cials operate the patronage networks to support the dictator s party. This e ect is only reversed if the opposition party is expected to win by a large margin. In that case, a pooling equilibrium emerges and appointed local o cials unambiguously support the reformist party. I test the empirical predictions of the model with a unique data set from Indonesia, which contains information on the electoral results for the rst and second post-suharto elections for over 30,000 villages. Indonesia is the ideal setting to explore the features highlighted by the model. The country is divided into two types of villages: desa and kelurahan. In desa the village head is elected by villagers, 4 while in kelurahan the village head is appointed by the district mayor. 5 3 The solution concept I use is Perfect Bayesian Equilibrium and then I focus on those equilibria that satisfy the Intuitive Criterion. 4 During the Suharto regime, elections for the village head in desa villages took place in a highly restricted set-up. Candidates were pre-screened and elections were nonpartisan. 5 Desa villages tend to be more rural while kelurahan tend to be more urban. Therefore, controlling for the di erences in the level of urbaness will be important for the econometric speci cation. Still, there is a good amount of overlap, since for historical reasons, some kelurahan were formed in quite rural areas. Also, the conversion of desa into kelurahan (as they became more urban) was stopped in 1992. Hence, I observe in some desa villages that are 3

By comparing the electoral behaviors of desa and kelurahan, within districts and when the main determinants of voting are controlled for, I aim to identify the di erences in the electoral outcome that can be attributed to the di erent selection method of their village heads. In the rst democratic election of Indonesia post-suharto, the electorate voted, simultaneously, for the national and district legislatures. 6 Since the designation rights of appointed village heads rested at the district level, they should have been particularly concerned about the electoral outcome at the district. This makes of the Indonesian case the ideal setting to explore whether the di erences in the electoral behavior of desa and kelurahan is related to the expected electoral outcome of the district in the way the theory predicts. The empirical results highlight that Suharto s party was, on average, 5 percentage points more likely to win in villages that had an appointed village head, relative to those that had an elected village head. This result is signi cant at the 1% level, robust to the inclusion of a broad set of controls and district xed e ects, and similar across di erent econometric methods (ordinary least squares and propensity score matching). Consistent with the model, this result is reversed for districts in which the main reformist party won by a large margin. In those districts, the reformist party is 4 percentage points more likely to win in villages with an appointed village head relative to those with an elected village head. This result is noteworthy since alternative hypotheses that rely on the existence of unobserved heterogeneity between these two types of villages, do not provide a satisfactory rationale for this empirical pattern. Finally, I examine the dynamic implications of the model with data from the second democratic election. The model predicts that in regions where a separating equilibrium emerged, village head turnover was high: since political leanings are truthfully revealed along the equilibrium path, when the winner of the rst democratic election takes o ce, she is able to detect her non-supporters and re them. In contrast, in places where the equilibrium is pooling, all appointed village heads exert the same level of e ort and consequently the composition of village heads remains unchanged. Notice that this leads to somewhat counterintuitive predictions for the second election. Support of appointed village heads for a given party should be higher in districts where that party won by a tight margin in the rst election, and lower if they won by a large margin (in the former case, the equilibrium was separating and non-supporters were red, while in the latter case, the equilibrium was pooling and non supporters are still in o ce). I provide some suggestive evidence that these mechanisms seem to be playing a role at the time of the second election. This paper is related to a number of di erent literatures. First, it relates to the literature that examines the speci c workings of new democracies in terms of their economic and political quite urban based on their observable characteristics, at the time of the rst democratic election in 1999. I provide further details in the empirical section. 6 The national and district legislature designated the head of the executive of the corresponding level of government. 4

outcomes. Some examples are Wantchekon (1999), Brender and Drazen (2005, 2008, 2009), and Keefer (2007). My paper contributes to this literature by providing microeconomic evidence of the workings of the rst and second democratic elections in Indonesia and by highlighting the importance of asymmetric information about political leanings in nascent democracies. Second, it relates to the political science and economics literature on democratic capture by the elite or other interest groups by means of vote buying, voter co-optation, patronage networks, and the use of force or its threat. Some examples are Gershenson and Grossman (2001), Bertocchi and Spagat (2001), Robinson and Verdier (2002), Acemoglu, Ticchi and Vindigni (2009), and Acemoglu, Robinson and Santos-Villagran (2009). My paper contributes to this literature by focusing on the role of local o cials and by providing evidence that, in the context of a regime change, unless the reformist parties are expected to be the clear winners of the rst democratic election, appointed local o cials will contribute to the persistence of the autocratic status quo. Third, it links to the literature that investigates the di erent incentives that elected versus appointed o cials face. See, for instance, Besley and Coate (2003), Maskin and Tirole (2004), and Alesina and Tabellini (2007, 2008). However, to my knowledge, this is the rst paper to point out that even non-elected o cials will have important electoral incentives in the elections for the politician or o cial that has decision rights over their appointment. Furthermore, I highlight that these incentives will be intensi ed when there is an additional motivation to signal certain political leanings. The rest of the paper proceeds as follows. Section 2 presents the model and derives its empirical predictions. Section 3 provides an overview of the Indonesian political structure and of the organization of the state. Section 4 describes the data. Section 5 explains the empirical strategy and discusses the main results. Section 6 presents the robustness checks that rule out competing explanations. In Section 7, I explore the dynamic implications of my model for the second democratic election. Finally, Section 8 presents the conclusions. 2 Model In this section, I develop a model to understand what incentives local o cials face at the onset of the rst democratic election and how they vary depending on their method of selection. For an easier comparison to the empirical part, I use the Indonesian terminology in the model. In particular, I refer to local o cials as village heads, some of which are appointed by the district mayor and others are elected in village level elections. However the model is, to a great extent, generalizable to other situations of two tiers of government in which designation rights of appointed local o cials rest on the upper tier, and local o cials have control over local patronage networks. 5

2.1 Setup Consider a district where two candidates are contesting the seat of district mayor. Each candidate belongs to either party A or B, with subscript m 2 fa; Bg denoting their party a liation, which is publicly known. Without loss of generality, let party A be the dictator s party and party B be the reformist party. This district is divided into N villages. In n of them the village head (he) is appointed by the mayor (she), while in the other N n villages the village head is elected by plurality rule elections held at the village level. The superscript v 2 fapp; elecg stands for the selection method of the village head in village v, with app and elec corresponding to appointment and election methods, respectively. Village heads have sympathies for one of the parties. The subscript t 2 fa; bg corresponds to a political sympathy towards party A or B, with population proportions of and 1, respectively. These political leanings are assumed to be private information, which is the most natural speci cation for the rst years of a democratic regime due to mainly two reasons. First, the previous non democratic regime probably repressed those that had views di erent from the dictators ideology. Thus, political leanings discrepant from the ideology of the regime might have been to a great extent hidden. Second, the events that lead to the fall of a nondemocratic regime and the beginning of a transitional period might considerably shape political attitudes, overall leading to a great deal of uncertainty about who supports whom, especially within the government administration. 7 District mayors have a preference for village heads that share their same political views, deriving additional utility G for each village head that is ideologically aligned to them. 8 Upon taking o ce, the district mayor has an opportunity to decide over the continuity in their positions of each appointed village heads. Let v 2 f0; 1g be the decision of the district mayor to dismiss or retain, respectively, the incumbent appointed village head of village v. If the mayor decides to dismiss him, she will have to incur in costs c, that capture the disutility of searching for a suitable candidate for the open position. From the point of view of the mayor, the bene t of taking that action is that she will 7 Notice that I do not need to assume imperfect information about the political leanings of elected village heads. Since they were selected into o ce by winning village level elections, we could expect that some information about their political views might have been disclosed at the time of those elections. The results of the model are the same regardless of the informational assumption of political leanings of village heads that are elected. What is important for the model is that the political leanings of appointed village heads are private information, which is a more plausible assumption for the reasons described above. 8 The parameter G might have a variety of interpretations. It can capture, in a reduced form way, the utility that the mayor derives from his preferred policies being implemented in the village. It can also account for the mayor s expectation of obtaining higher electoral support in subsequent elections from a village in which the village head is a sympathizer of her same party. 6

be able to appoint one of her cronies as village head that she knows for sure is aligned to her. 9 Overall, the utility that a district mayor of party A and B, respectively, derives from a village with an appointed village head is: V app A (; t) = G1 ft=ag + (1 )[G c] (1) V app B (; t) = G1 ft=bg + (1 )[G c] (2) where 1 ft=jg is a dummy that takes value 1 if the village head is a party j supporter (j 2 fa; bg), and 0 otherwise, G is the additional utility that the district mayor obtains when the village head is aligned to her, c are the costs incurred if the incumbent village head is dismissed (which satisfy G > c), and takes value 1 if the mayor decides to retain the village head and 0 otherwise. Similarly, the utility that a district mayor of party A and B, respectively, derives from a village with an elected village head is: V elec A (t) = G1 ft=ag (3) V elec B (t) = G1 ft=bg (4) Notice that the only di erence between the utility mayors derive from villages with an appointed village head versus those with an elected village head, is that in the latter case the mayor can not dismiss the village head. I now de ne the preferences of village heads. If a village head is able to keep his position, he obtains rents R from being in o ce, whereas if he is red, he obtains his reservation utility U. Throughout the paper, I will focus on cases in which village heads are interested in keeping their positions, i.e. R > U. During the mayoral electoral campaign, village heads can exert e ort to persuade voters in their village to vote for party A or B. The possibility of in uencing voter behavior is particularly plausible in the context of the rst democratic election. Local patronage networks and other cooption mechanism, which are the cornerstone of many nondemocratic regimes, are likely to still be present at the onset of the rst democratic election. To better understand the patterns of political support in new democracies, in this model each village head will decide which candidate to favor when operating the patronage networks under his control. Let e 2 R be the level of e ort that a particular village head decides to exert in order to persuade voters to vote for party A: 9 Notice that I am assuming that during the democratic period, each party has a group of strong supporters that are committed to the party and there is no uncertainty about their political leanings. However, this group might be small and the costs c captures the opportunity cost of appointing them as village heads and not to alternative jobs. In contrast, during the nondemocratic regime, anyone that wanted to be a village head had to pretend to share the same ideology as the dictator. Hence, giving candidates for village heads strong incentives to hide their real political leanings. The results of the model will still hold if the technology to identify supporters is only slightly better during the democratic period than in the nondemocratic regime, which seems a plausible assumption. 7

thus, positive (negative) values of e will improve the electoral prospects of party A (B). Exerting e ort will be costly for village heads, captured by a twice continuously di erentiable cost function C(:) : R +! R +, de ned over the absolute value of e ort satisfying C(0) = 0, C 0 (j:j) > 0, C 00 (j:j) > 0. 10 Exerting e ort will be more costly for a village head when it favors his least preferred candidate. To capture this, a parameter or (satisfying > ) will multiply the cost function above, depending on the direction of the e ort exerted. 11 Therefore, the utility of an appointed village head that is a party A or party B supporter, respectively, is: U app a (e; ) = R + (1 )U (1 fe<0g + 1 fe>0g )C(jej) (5) U app b (e; ) = R + (1 )U (1 fe<0g + 1 fe>0g )C(jej) (6) where takes value 1 if the village head keeps his position and 0 otherwise, 1 fe<0g and 1 fe>0g are indicator functions that take value 1 if e ort, e, is negative or positive, respectively, and C(:) is the cost of e ort. Similarly, the utility of an elected village head that is a party A or party B supporter, respectively, is: U elec a (e) = R (1 fe<0g + 1 fe>0g )C(jej) (7) U elec b (e) = R (1 fe<0g + 1 fe>0g )C(jej) (8) Notice that the only di erence in the preferences of appointed and elected village heads is that the latter ones cannot be red. Hence, the utility of elected village heads does not depend on which mayor wins the election. This will lead to important di erences between the e ort exerted by elected and appointed village heads. Finally, I specify how the e ort of village heads a ects the electoral outcome. I assume there is common knowledge about the share of the population that has a preference towards party A and denote that proportion by. There are two other factors that can a ect the electoral outcome. 1 First, a valence shock uniformly distributed in the interval [ 2 ; 1 2 ], which captures the unexpected component of the relative popularity of candidate A with respect to candidate B. is a parameter that measures the density of the valence shock distribution, hence, it is inversely related to the variance of the shock. Second, the sum of e orts of village heads can also have an impact on the electoral outcome. Therefore, the realized vote share of candidate A can be speci ed in the 10 Throughout the paper, in order to minimize notation, I will omit the notation for absolute value from the cost function, whenever it is obvious from the context that we are considering positive levels of e ort. 11 The introduction of partisan preferences through the cost of e ort leads to similar results as introducing an additional payo for village heads if their preferred party wins the election. However, the current speci cation permits a cleaner interpretation of the di erences in the e ort exerted by elected and appointed o cials. As it will be discussed later, e is interpreted as the part of e ort that comes motivated by the di erent selection mechanism. 8

following way: e = + + g(e) (9) P where E is the sum of the e ort levels of all village heads in the district (i.e., E = N e i ), g(:) is a twice continuously di erentiable function satisfying g(0) = 0; @g(e) @E e ort a ects the realized vote shares. i=1 > 0; which captures how total E ort of village heads is assumed to be observable to both candidates for mayor. 12 One possible interpretation of this speci cation is that there is perfect information about the preferences of the median voter in each village and any deviation from that is attributed to the e ort exerted by its village head. Also, village heads themselves might have an incentive to make their e ort level observable and therefore, might be vocal about it. The following expression shows the probability that candidate A wins the mayoral election as a function of total e ort level: p(e) = Prob [e 1 2 ] = [ + g(e) 1 2 ] + 1 2 (10) I now proceed to summarize the timing of events. 1. Taking into account, every village head chooses a level of e ort e i 2 R. 2. The electoral outcome is realized and the level of e ort that village heads exerted is observed. The candidate for mayor that obtains the largest vote share takes o ce. 3. The new mayor decides whether to retain or dismiss each appointed village head i 2 f0; 1g. 4. Payo s are distributed and the game ends. 2.2 Characterization of Equilibria In this section, I de ne the solution concept and characterize the set of equilibria. An equilibrium consists on a pair of strategies for the two candidates for mayor, a set of strategies regarding e ort decisions for appointed and elected village heads, and a set of beliefs about village head types. I rst describe the optimal e ort level that elected village heads exert in any equilibrium. Since the continuity of elected village heads in their positions neither depends on which mayor wins the election, nor on the strategies mayors play, it is straightforward to see that elected village heads do not nd optimal to exert e ort. The following proposition summarizes this result. 12 An extension of the model in which e ort levels are observed with noise, will be available in the next version of this paper. The main intuitions provided by this model still hold. However, the nature of the pooling equilibrium changes slightly, since one of the type of players will play a pure strategy whereas the other will play a mixed strategy. 9

Proposition 1. In any equilibrium, elected village heads exert zero e ort regardless of their political leanings e elec a = e elec b = 0 Elected village heads keep their position either if mayor A or mayor B wins the election and at the end of the game they receive payo R with certainty. Proof. The level of e ort that maximizes the utility of an elected village head of type a, given by (7), is e elec a = 0. Similarly, the optimal e ort of type b elected village head, given his preferences de ned by (8), is e elec b = 0. Since even exerting no e ort they can keep their positions as village heads, they can not increase their payo s by choosing any other level of e ort. This result should not be interpreted as predicting that elected village heads will never exert e ort to support one party or another. They might derive some intrinsic utility from the victory of a particular candidate. Also, district mayors might distribute additional funds to village heads that are aligned to them. In these scenarios, the elected village head might nd optimal to exert some amount of e ort during the mayoral electoral campaign. However, there is no reason why these additional incentives should not be also present for appointed village heads. One of the objectives of this model is to isolate the level of e ort that comes motivated by the di erent selection mechanism, and that is how we should interpret e. Let us now turn to the game de ned by appointed village heads and the two potential candidates for mayor. Notice that e ort exerted by appointed village heads has a twofold motivation: rst, it can potentially a ect the outcome of the election and second, it can signal a particular political a liation. When analyzing the optimal behavior of an appointed village head, the setting constitutes a dynamic game of incomplete information, more speci cally a signaling game between the village head and the two potential candidates for mayor. The solution concept I use to solve this game is Perfect Bayesian Equilibrium and I re ne the set of equilibria using the Intuitive Criterion. 2.2.1 Solution Concept De nition 1. A Perfect Bayesian Equilibrium (PBE) of this game consists on a set of optimal strategies for both candidates for mayor m(e i ) 2 f0; 1g m 2 fa; Bg, a set of optimal strategies for each appointed village head e i (t) 2 R t 2 fa; bg, and a set of posterior beliefs (tje i) such that e i (t) 2 arg max e i n o m(e i ) 2 arg max (tje i )Vm app (; t) t (11) fp(e i +e i )U app t (e i ; A(e i )) + (1-p(E i +e i )) U app t (e i ; B(e i ))g (12) where (tje i ) is derived from the prior (population shares), e i, and e i (t) using Bayes rule (when applicable), Vm app (; t) m 2 fa; Bg are de ned by (1) and (2) respectively, U app (e; ) t 2 fa; bg are t 10

de ned by (5) and (6) respectively, p(:) is de ned by (10), and E i is the aggregate e ort level of all village heads other than i, i.e., E i = P j6=ie j. In the rest of this section, I characterize the set of equilibria of this game, focusing on the interaction of appointed village heads and the two candidates for mayors. 13 Therefore, in order to minimize the use of notation I will drop the superscript app, which stands for appointed village head. 2.2.2 Mayor Optimization Problem Upon taking o ce the new mayor decides, based on the observed levels of e ort, whether to keep or dismiss each appointed village head in her jurisdiction. She will decide to keep a particular village head as long as the expected utility from doing so is higher than the expected utility of dismissing him. By noting that the utility function of mayor A is given by (1), it is straightforward to see that she will nd optimal to keep a village head that exerts e ort e as long as the following holds: (t = aje)g > G c (13) where (t = aje) is the posterior probability that a village head is type a given that he exerted e ort level e. Mayors derive this posterior probability using Bayes rule when applicable. 14 Similarly, if the candidate for mayor of party B takes o ce, she will keep the village head as long as [1 (t = aje)] G > G c. Therefore, notice that their decisions depend on their assessment of how likely is the village head to be politically aligned to them, and on the relative bene ts of an aligned village head relative to the costs of ring. 15 2.2.3 Pooling Equilibria Next, I analyze the set of pooling PBE of this game, in which both types of village heads exert the same level of e ort e (t) = be for t 2 fa; bg. Notice that, mayors will not be able to update their prior along the equilibrium path. Consequently, mayors will equate the posterior probability of a village head being of a particular type to the corresponding population share of that type i.e., (t = ajbe) =. By plugging this probability in the optimal decision rule of the mayor A given 13 Since elected village heads always exert zero e ort they do not play any role in this game. 14 In this setup, the Bayes rule is (t = aje) = appointed village head of type t exerts level of e ort e. P (ejt=a) P (ejt=a)+p (ejt=b)(1, where P (ejt) is the probability that an ) If e is an action taken along the equilibrium path, this probability is determined by the strategies played in equilibrium. However, if e is not played along the equilibrium path, the Bayes rule does not pin down the posterior probability. 15 Given the timing of events, mayors cannot commit to implement any strategy di erent than their optimal one upon being elected. Otherwise, they might nd optimal to o er a more sophisticated contract to village heads during the electoral campaign. This is why the preferences of mayors that are relevant are those at an interim stage, i.e., after being elected. 11

by (13), it is straightforward to see that mayor A will nd pro table to keep a village head that exerted e ort be as long as > G c G. Similarly, mayor B will keep a village head that exerted e ort be if 1 > G c. Hence, depending on how the proportion of each type of village head relates to the ratio G G c G di erent strategies of mayors can be sustained in equilibrium. In this subsection I examine the following set of parameters. CASE 1. > G c G > 1 (14) In this case, the proportion of type a village heads is particularly high. As I describe below, pooling equilibria will be sustained in this set of parameters if the underlying support for party A is high enough. In the Appendix I discuss the opposite case, in which the proportion of type b village heads is high and pooling equilibria emerge provided that the underlying support of party B is high enough. Since the underlying support of a party in the population is likely to be positively correlated with the proportion of appointed village heads that are sympathizers of that party, these are the most relevant parameter sets in which pooling PBE might emerge. Therefore, in the rest of the paper I will focus on pooling equilibria for emerges for these two cases. 16 Consider the following strategies and beliefs as a candidate for PBE of this game for Case 1: 17;18 A(e) = ( 1 if e = be 0 if e 6= be B(e) = ( 0 if e = be 1 if e 6= be e i (t) = be 0 for t 2 fa; bg (t = aje = be) = (t = aje 6= be) = 0 (15) Therefore, along the equilibrium path, if mayor A wins the election, she keeps all the appointed village heads, whereas if mayor B is elected, she res all of them. Notice that these strategies are sustained because the proportion of type a village heads is high, relative to the proportion of type b village heads. Since type b village heads have higher costs of exerting positive e ort, they are the 16 For completion, the set of pooling equilibria that might emerge for other sets of parameters is also analyzed in the Appendix. 17 Notice that in equilibrium all village heads of a particular type will exert the same level of e ort. This result is derived from the symmetry of their optimization problems and will be common to all equilibria described in this paper. In order to minimize notation, oftentimes I will omit the i subscript. However, when checking for deviations from the equilibrium path I consider the deviation of a single individual of a particular type, holding constant the actions of any other village head of either type. 18 I focus on the set of equilibria in which village heads e ort is aimed at supporting candidate A, i.e. be 0. There can be pooling PBE with associated be < 0, but these peculiar equilibria in which village heads support party B but only party A hires them, do not satisfy the Intuitive Criterion and I do not discuss them further. 12

most likely ones to deviate from their equilibrium strategy. Let e 0 b be the most pro table deviation of a particular type b village head, given that every other village head is exerting e ort be. 19 e 0 b = arg max f[1 p([n-1] be+e)] (R U) C(jej)g (16) e0 Village head type b will not nd pro table to deviate as long as: p(e )(R U) C(be) 1 p(e 0 ) (R U) C(je 0 bj) (17) 1 2 [R-U] [R-U] 1 g(e ) g(e 0 ) + C(be) C(je 0 b j) (18) where E = nbe is total e ort in equilibrium, E 0 = [n-1] be+e 0 b is total e ort if a village head type b deviates, and the last inequality follows by plugging in the expression for the probability that party A wins the election, given by (10). A number of features from the above expressions are worth noticing. First, the stronger is the underlying support for party A in a district (higher ) the more likely is this equilibrium to exist. This result is actually very intuitive: when the election is expected to be very lopsided, all the appointed village heads have a strong incentive to pretend to be supporters of the likely winner. Second, the lower the level of e ort required to exert in equilibrium, be, the more likely is this equilibrium to exist. A low required e ort minimizes the incentives that type b has to deviate. Third, the smaller are the di erences in costs of e ort, the more likely is this equilibrium to exist. Hence, there can not be strong partisan preferences among village heads, otherwise it would be very costly for village heads to support their least preferred candidate, giving them strong incentives to deviate. Notice that the set of PBE is very large, since there can be in nitely many levels of e ort that satisfy inequality (18). However, some of these pooling equilibria are sustained by out of equilibrium beliefs that are not always reasonable. A standard practice in this type of games is to apply some re nement to the equilibrium concept. In particular, I consider the Intuitive Criterion rst formalized by Cho and Kreps (1987). The application of this re nement eliminates many pooling PBE. 20 However, the following level of e ort is associated to a PBE that satis es the Intuitive Criterion. 21 @g(ne a) @E [R U] = C 0 (e a) (19) This e ort level maximizes the expected payo s of type a and there is no deviation that could 19 The optimal deviation necessarily satis es e 0 b 0, since deviating to e 0 > 0 is always dominated by deviating to e 0 = 0: both lead to the same actions of mayors, but in the latter case the village head saves the cost of e ort. 20 See the proof of Proposition 2 in the Appendix for the de nition of the Intuitive Criterion and its application to the current setup. 21 This is under the assumption that there is an interior solution, i.e. @g((n 1)e a ) @E [R U] > C 0 (0). 13

make type a better o, for any possible out of equilibrium beliefs. 22 Therefore, if there was some deviation, mayors should deduce the village head is type b. Hence, type b could potentially reveal his type by undertaking certain deviations. However, it would never be in his best interest to do so. Upon revealing his type, the best response of mayor A would be to dismiss him and only mayor B would be willing to keep him. Since inequality (17) holds, this is not pro table for type b. In other words, type a is getting his maximum payo and could not reveal his type by deviating to an alternative level of e ort. On the contrary, type b could reveal his type by undertaking certain deviations but he would never want to do so. Consequently, the pooling equilibrium described above satis es the Intuitive Criterion. Notice that there might be other PBE that also satisfy the Intuitive Criterion. 23 However, the equilibrium above is the only one that also satis es stronger equilibrium re nements, such as Universal Divinity (Banks and Sobel (1987)). 24 The following proposition summarizes these results. Proposition 2. If condition > G c G > 1 is satis ed, for each be 0 such that inequality (17) holds, the set of strategies and beliefs speci ed in (15) constitutes a pooling Perfect Bayesian Equilibrium of this game. In this equilibrium, all appointed village heads exert e ort be and along the equilibrium path both keep their positions if candidate for mayor A wins the election and are dismissed otherwise. The PBE associated to level e ort e a de ned by (19) satis es the Intuitive Criterion. Proof. In the Appendix. 2.2.4 Separating Equilibria Let us now turn to describe the set of separating equilibria in which each type of village head takes an action perfectly distinguishable from the action of the other type. Hence, along the equilibrium path, types will be truthfully revealed and mayors are able to identify and only retain those village heads that are aligned to them. Let n a (n b ) be the number of appointed village heads that are type a (type b). 25 Consider the following set of strategies and beliefs as a candidate for separating PBE 22 E ort level e a is de ned such that, conditional on all other appointed village heads exerting e ort level e a, the optimal deviation of type a is exactly to e ort level e a. In particular, e a = argmax fp((n-1)e a+e)(r-u) C(e)g. This ensures type a does not have a pro table deviation for any out of equilibrium beliefs. Also notice that this optimal level of e ort is maximizing the expected utility of the village head. Hence, village heads will take into account the impact of their e ort on the electoral outcome. In other words, they not only have the signaling motivation of e ort, but also some electoral motivation. For further discussion see the proof of Proposition 2 in the Appendix. 23 For instance, consider a PBE associated to a very large level of e ort. Both types would be better o by deviating (conditional on mayors revising their out of equilibrium beliefs). But since both types would bene t from doing so, they can not reveal their type undertaking such deviations. 24 A formal proof will be provided in the next version of the paper. The main intuition of why the divinity criterion eliminates PBE other than the one associated to e a, is that in those other equilibria, type a will always be more likely to deviate be to deviate to e a than type b. 25 Therefore, the proportion of appointed village heads that are type a is = na n and type b is 1 = n b n. 14

of this game. A(e) = ( 1 if e 0 0 if e < 0 B(e) = ( 0 if e 0 1 if e < 0 e i (t = a) = e s e i (t = b) = e s ( 1 if e 0 (t = aje) = 0 if e < 0 (20) where e s is implicitly de ned by @g([n a -n b ] e s ) @E [R U] = C 0 (e s ) (21) Notice that given the speci ed out of equilibrium beliefs, mayor A will retain any village head that exerts a positive level of e ort. Therefore, the action that type a takes in equilibrium needs to maximize his expected payo s when the e ort of the rest of village heads is taken as given. Similarly for type b. This is the case when they exert the level of e ort de ned by condition (21). The following additional conditions ensure that type a does not want to pretend to be type b by deviating to a negative level of e ort, and vice versa. p(e s )(R U) C(e s ) 1 p E 0 a (R U) C(jea j) (22) 1 (R-U) 1 g(e s ) g(e 0 2 (R-U) a) C(je a j) + C(e s ) (23) where E s = [n a -n b ] e s is total e ort in equilibrium and E 0 a = [n a -n b -1] e s +e a is the total e ort if a type a village head deviates, and e a is type a s optimal deviation de ned by e a = arg max f[1 p([n a -n b -1] e s +e)] (R U) C(jej)g (24) e<0 Similarly, type b will not have incentives to deviate to positive levels of e ort if the following holds: 1 (R-U) 1 g(e s ) g(eb 0 2 (R-U) ) + C(e b ) C(e s ) (25) where E 0 b = [n a-n b +1] e s +e b is the total e ort that emerges if type b village head deviates, and e b is type s b optimal deviation de ned by e b = arg max fp ([n a -n b +1] e s +e) (R U) C(e)g (26) e0 The following proposition summarizes these results. 15

Proposition 3. If conditions (23) and (25) hold, the set of strategies and beliefs speci ed in (20) constitutes a separating Perfect Bayesian Equilibrium of this game. In this equilibrium type a appointed village heads exert e ort e s as de ned by (21), type b appointed village heads of exert e ort e s, mayor A only retains appointed village heads that exerted e ort e s, and mayor B only retains appointed village heads that exerted e ort e s. This equilibrium satis es the Intuitive Criterion. Proof. In the Appendix. A number of features are worth noticing from this proposition. First,, the underlying strength of party A in the district, needs to take intermediate values for this equilibrium to exist. In other words, separating equilibria will emerge when the election is expected to be contested. Intuitively, both candidates for mayor need to have some chance of winning the election. Otherwise, village heads would have strong incentives to support their least preferred candidate if she is very likely to win. Second, notice that for separating equilibria to exist, the di erence in costs of supporting the most preferred candidate versus the least preferred one, i.e., needs to be high enough. Thus, there needs to be strong enough partisan di erences among village head to sustain a separating equilibrium. Otherwise, village heads will have incentives to deviate to support their least preferred candidate even if she is only slightly more likely to win. 2.3 Aggregate E ects With the objective of obtaining empirically testable predictions, in this section I investigate how the aggregate level of e ort of appointed village heads depends on the characteristics of each district. The following proposition summarizes the previous results and describes what will be the level of aggregate e ort exerted in each district, when focusing on equilibria which satisfy the Intuitive Criterion: Proposition 4. 1. If condition 1 > G c G > holds and is low enough (inequality (31) is satis ed), a pooling PBE emerges in which all appointed village heads exert e ort to support party B. Total e ort in the district will be ne a < 0 where e a is de ned by (19). 26 2. If takes intermediate values (inequalities (23) and (25) hold), a separating PBE emerges. Total e ort in the district will be n a e s n b e s where e s is de ned by (21). 3. If condition > G c G > 1 holds and is high enough (inequality (17) is satis ed), a pooling PBE emerges in which all appointed village heads exert e ort to support party A. Total e ort in the district will be ne a > 0 where e a is de ned by (19). 26 See the Appendix for the discussion of pooling PBE for this set of parameters. 16

Proof. Statement 1 follows from Proposition 5 in the Appendix. Statements 2 and 3 follow from Propositions 3 and 2, respectively. As the proposition above highlights,, the relative strength of party A in a district, plays a crucial role in determining whether a pooling or a separating equilibrium emerges. However, can also a ect the aggregate level of e ort by other channels. First, the productivity of e ort is likely to be higher when the election is expected to be contested, since it should be easier to persuade citizens to vote for a particular candidate when there is a higher probability that their vote is going to be pivotal. To account for this let us consider that the function g(:), which captures the how e ort a ects the vote shares, also depends on, i.e. g(e; ). In particular let us assume that it takes the following form g(e; ) = E h(), where h() is maximized when = 1 2. Under this assumption, the individual level of e ort exerted in equilibrium, as de ned by (19) and (21) will be a function of which has an inverse U-shape form maximized at = 1 2. A second feature that might change across districts is the ideological composition of village heads. We would expect and to be positively correlated since districts in which one of the parties has strong popular support, might also have a substantial proportion of village heads that are sympathizers of that party. After all, village heads are a subgroup of the population. Still appointed village heads are probably not a random sample, since they were appointed by the dictator s party (party A) during the nondemocratic regime. Therefore, it is likely that at any district the proportion of village heads that are party A sympathizers outnumbers their population counterpart. Figure 1 summarizes these two additional assumptions. On the left panel, the optimal individual e ort is plotted as a function of the underlying strength of party A, i.e.. Similarly, the panel on the right shows how the proportion of village heads that are party A sympathizers might depend on. The introduction of these two assumptions permits us to obtain speci c predictions about how aggregate e ort di ers across regions. As I will discuss in the empirical section, the data exhibits a heterogenous pattern substantially similar to the one predicted when these two additional assumptions are established, which is reassuring of the assumptions introduced. The following result summarizes these assumptions and describes their implications. Result 1. Under the following additional assumptions 1. g(e; ) = E h() where h 0 (:) > 0 if < 1 2, h0 (:) < 0 if > 1 2 and h00 (:) < 0. 2. () satis es 0 (:) > 0 and () > 8 the aggregate e ort described in Proposition 4 has a pattern as described in Figure 2. 27 27 See section 9.4. in the Appendix for a more detailed discussion on this Result. When these additional assumptions 17

Figure 2 plots how aggregate e ort depends on, as described by Proposition 4, when these two assumptions are taken into account. A number of features are worth emphasizing. First, notice that, even when the equilibrium is separating, the high proportion of appointed village heads that are supporters of the dictator s party (party A) can lead to aggregate levels of e ort that are positive. Only if the reformist party is expected to win by a large margin we would expect to obtain an unambiguous negative aggregate e ect (i.e., appointed village heads giving their support to party B). Second, the di erence in aggregate e ort between pooling and separating equilibria might not be too high. Even though in pooling equilibria all village heads exert e ort in the same direction, they might be coordinating in low levels of e ort. Therefore, I do not expect to nd a discontinuity in the data that would enable me to test for the type of equilibria. 2.4 Summary of Empirical Predictions In this subsection I summarize the empirical predictions of the theory described. 1. In most regions, the aggregate level of e ort of appointed village heads favors the dictator s party, especially in regions where it has a strong underlying support in the population. 2. The e ect is reversed in regions where the reformist party is expected to win by a large margin. In those regions appointed village heads support the reformist party. 3. If assumptions 1 and 2 of Result 1 are satis ed, the aggregate level of e ort, as a function of the underlying strength of the dictator s party, has an heterogenous pattern as displayed in Figure 2. In Section 5 of this paper, I test these empirical predictions with data from the Indonesian rst democratic election post-suharto. I compare the electoral outcome between villages with an elected village head and those with an appointed village head, within districts and when the main determinants of voting behavior are controlled for. By focusing on this comparison, I attempt to capture the di erences in their voting patterns that can be attributed to the behavior of their village heads. The empirical results corroborate, to a great extent, the predictions of the model. In particular, the data re ects an heterogenous pattern similar to the one described in Figure 2. This nding is particularly noteworthy, because alternative explanations that rely on the existence of unobserved heterogeneity between these two types of villages can not account for this pattern. Finally, I will examine the consequences of the pooling and separating equilibria in the second election and provide some suggestive evidence in Section 7. are introduced, there are no longer closed form solutions for the thresholds of separating and pooling equilibria. As long as n is large enough and function h(:) is not too sensitive to changes in, the thresholds are well behaved. 18